PAG-GNNA: Advanced Graph Neural Network Approach for 3D Human Pose Estimation and Action Recognition

In the past few years, computer vision has improved a lot especially in human pose estimation and action recognition. Neverth eless, these tasks are not easy as the combination of skeletal and visual features is rather complicated. Even though deep neural networks have demonstrated good performances, problems like high error rates and less accuracy remain. To overcome these difficulties, the current paper will suggest a new Pose and Action Guided Graph Neural Network Attention (PAG-GNNA) video-based human action recognition framework. The suggested approach uses a multi-modal, multi-stream design, which incorporates pose estimation and object detection modules, followed by a feature embedding and classification phase. The skeleton stream generates movement and poses data in the feature embedding module, which is integrated with learnable graph representations to boost feature robustness and discriminative power. Experimental results on the UCF101 and HACS datas ets show that the proposed approach has better performance and gets an accuracy of 96.3% and 96.2%, respectively, which outperforms the existing methods.